CN111096830A

CN111096830A - An exoskeleton gait prediction method based on LightGBM

Info

Publication number: CN111096830A
Application number: CN201911384974.3A
Authority: CN
Inventors: 孔万增; 王伟富; 宋国明; 王雪岩
Original assignee: Hangzhou Dianzi University
Current assignee: Hangzhou Dianzi University
Priority date: 2019-12-28
Filing date: 2019-12-28
Publication date: 2020-05-05
Anticipated expiration: 2039-12-28
Also published as: CN111096830B

Abstract

The invention discloses an exoskeleton gait prediction method based on LightGBM. The invention extracts the acceleration data of the lower limb joints for offline analysis through the analysis of human body kinematics, and then solves the joint control target trajectory. The invention proposes a new lower limb motion gait trajectory prediction method, which can be applied to the lower limb exoskeleton control, innovatively realizes the parallel structure input to the LightGBM, and predicts the gait trajectory of the continuous target value of the lower limb joints, and the prediction method It has strong accuracy and reduces training time.

Description

Exoskeleton gait prediction method based on LightGBM

Technical Field

The invention belongs to the field of human-computer cooperative motion control of lower limb exoskeleton, and relates to a motion gait trajectory prediction method based on a LightGradient Boosting Machine (LIGHTGM).

Background

The lower limb exoskeleton robot is a typical man-machine integrated system worn outside the lower limbs of a user, integrates the detection, control, information fusion and other robot technologies, combines the intelligence of the user with the physical strength of the robot, and provides power to assist the user in moving. In the civil field, the exoskeleton robot can help the old to normally act. In the aspect of the medical field, the exoskeleton robot can assist the disabled in normal life and greatly reduce the working pressure of medical staff. In the aspect of military field, the exoskeleton robot can improve the rescue efficiency of a battlefield and help more injured people, and the exoskeleton robot plays a great role in various fields, so that the exoskeleton robot has a very wide development prospect.

At present, two systems are used for inputting to predict the gait track of the exoskeleton robot, firstly, a sensing type sensor is used for measuring human body biological signals such as electroencephalogram (EEG) or Electromyogram (EMG) and the like to be used as the input of an exoskeleton control system, although the biological signals are ahead of the human body movement and can solve the problem of movement lag, the EEG and EMG signals are unstable and are easily interfered by static electricity, sweat and the like. And secondly, physical sensors such as angle/angular velocity, force/moment and the like are utilized to acquire human kinematics or dynamic data in real time to be used as the input of the exoskeleton control system. For the gait trajectory prediction problem, the traditional method comprises a Support Vector Machine (SVM), Kalman filtering and the like, the traditional method has overlarge calculated amount and large memory occupation and is easy to generate overfitting, and the continuous target value prediction of the gait trajectory still needs to be further improved.

Therefore, it is necessary to provide a lower extremity exoskeleton gait prediction method capable of improving the accuracy of continuous target value prediction of gait tracks, improving track smoothness and reducing the calculation amount.

Disclosure of Invention

Aiming at the defects of the prior art, the invention provides a gait track prediction method based on an improved LightGBM, which extracts acceleration data of lower limb joints for off-line analysis through human body kinematics analysis, further calculates a joint control target track, and finally controls the motion through the track by a control system to realize human body gait motion of lower limb exoskeleton.

In order to achieve the purpose, the technical scheme of the invention comprises the following specific contents:

a LightGBM-based exoskeleton gait prediction method, the method comprising the steps of:

step (1), data acquisition stage

Respectively arranging 5 IMU sensors on the waist, the left thigh, the left calf, the right thigh and the right calf of a user; acquiring x and y axial acceleration values of 5 IMU sensors in real time;

step (2), data analysis and noise elimination smoothing processing

Constructing a matrix consisting of 10 xk acceleration components according to the x and y axial acceleration values of the 5 IMU sensors in the step (1), wherein the matrix is defined as:

R＝[V₁,V₂,…,V_i,…,V₁₀]^Tt represents the transpose of the matrix

Wherein

Representing the acceleration value of the ith acceleration component at the moment t, wherein K is the total number of data sampling points;

to V_iPerforming moving average filtering and de-noising processing

Has a filtered value of

Then:

wherein L is the number of the original acceleration values obtained during the moving average filtering processing, and is an odd number, and M is (L-1)/2;

according to the formula (2.1), V is finally obtained_iNoise-canceling data at all times t (t ═ 1,2, …, K):

step (3) acquiring real-time flexion and extension angles in gait data

According to the noise elimination data processed in the step (2), taking the left lower limb as an example, the hip joint flexion and extension angle of the left lower limb at a certain time t

And the flexion-extension angle of the knee joint

The following equations (3.1) and (3.2) respectively yield:

wherein a is_x1,a_y1The acceleration components, a, of the left thigh IMU sensor at the time t after the processing in step 2 are respectively_x2,a_x2Acceleration components of the left and right crus IMU sensors at the time t after the processing in the step 2 are respectively obtained;

finally obtaining the flexion and extension angle vector α of the hip joint and the knee joint of the left lower limb at all K moments^{Left side of}And β^{Left side of}：

Obtaining flexion and extension angle vectors of the hip joint and the knee joint of the right lower limb in the same way;

step (4), continuous target value real-time prediction of improved LightGBM method

At the hip flexion and extension angle α of the left lower limb^{Left side of}For example, the following steps are carried out:

4.1 predictive model training

Training a prediction model of the hip joint flexion and extension angle of the left lower limb by taking the vector data of the hip joint flexion and extension angle of the left lower limb obtained in the step 3 as a training set;

first using a sliding window from α^{Left side of}Extracting angle values from the left lower limb hip joint flexion and extension angle matrix A_{α left side}As a feature matrix for training the prediction model, see formula 4.1;

wherein w is the window width of the sliding window, N is the number of the sliding windows, the predicted length pLen is taken as the step length of the sliding window, and pLen is more than 1; to ensure better construction of the target matrix, K- [ (N-1) × pLen + w needs to be satisfied]Not less than pLen, i.e. ensuring α^{Left side of}Finally, enough angle values are provided to construct predicted values of the pLen predicted lengths;

then according to the predicted length and α^{Left side of}Constructing a matrix D according to the future-time flexion-extension angle value of the current sliding window_{α left side}As a prediction target value matrix for training a prediction model, see formula 4.2;

wherein

A predicted value vector corresponding to the ith future moment of the hip joint flexion and extension angle of the left lower limb is obtained;

and (3) constructing a feature matrix of the left lower limb knee joint according to the formulas (4.1) and (4.2) and substituting the feature matrix into a formula (4.3) to obtain pLen prediction models:

where T () is the LightGBM training function,

a prediction model (function) corresponding to the ith future moment after the hip joint flexion and extension angle of the left lower limb is trained;

4.2 parallel prediction of prediction models

The feature vector in the current latest sliding window of the hip joint of the left lower limb is

The prediction of the hip joint flexion and extension angles of the left lower limb at future pLen moments is realized by connecting pLen trained prediction models in a parallel structure:

wherein,

predicting the hip joint flexion and extension angle of the left lower limb at the ith moment in the future;

the left lower limb hip joint flexion and extension angle prediction vector P can be obtained from the formula (4.4)^{Left side of}：

Obtaining a flexion-extension angle parallel prediction model of the knee joint of the left lower limb, the hip joint of the right lower limb and the knee joint in the same way;

and (5) predicting the vector according to the step 4 to realize the gait prediction track.

The invention has the beneficial effects that:

the invention provides a brand-new lower limb movement gait track prediction method which can be applied to lower limb exoskeleton control, innovatively realizes parallel structure input to LightGBM, predicts the gait track of a continuous target value of a lower limb joint, has high accuracy and reduces training time.

Drawings

FIG. 1 is a graph of four raw data versus noise-canceled data; wherein (a) left calf acceleration x; (b) left calf acceleration y; (c) left thigh acceleration x; (d) left thigh acceleration y;

FIG. 2 is a diagram of a model of a lower limb structure of a human body;

FIG. 3 is a comparison of three Kalman filtering, XGBOST, and LightGBM methods, where (a) is RMSE, (a) is SC, and (c) is training time;

fig. 4(a) and (b) are comparison results of verification of hip joint and knee joint of left lower limb, respectively.

Detailed Description

In order to make the objects, technical solutions and points of the present invention clearer, embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

A LightGBM-based exoskeleton gait prediction method comprises the following steps:

1. data acquisition phase

Respectively arranging 5 IMU sensors on the waist, the left thigh, the left calf, the right thigh and the right calf of a user; the lower limb model diagram shown in fig. 2 can be simplified according to the structure of the lower limb of the human body, 5 black points represent the positions of 5 IMU sensors, each IMU sensor provides an acceleration component on a two-dimensional plane corresponding to each leg in the walking process of the lower limb of the human body, the hip joint takes the anticlockwise direction as the positive direction, and the knee joint takes the clockwise direction as the positive direction, so that the x-axis and y-axis acceleration values of the 5 IMU sensors are obtained in real time.

The acquisition equipment is 5 IMU sensors which are respectively arranged on the right shank, the right thigh, the left shank, the left thigh and the waist, the sampling rate is 100Hz, and the testee walks linearly at the pace of 2 km/h.

2. Data analysis and noise cancellation smoothing

R＝[V₁,V₂,…,V_i,…,V₁₀]^Tt represents the transpose of the matrix

Wherein

to V_iPerforming moving average filtering and de-noising processing

Has a filtered value of

Then:

in the actual operation process, if the calculation is carried out according to the formula (1)

Then N summation operations are required at each time, and the time complexity is high. In general, the moving average filter can be implemented by a recursive algorithm.

Applying the partial acceleration component data to the MAF algorithm, four raw data plots are plotted versus noise-canceled data, as shown in fig. 1.

3. Real-time joint angle calculation of gait data

And the flexion-extension angle of the knee joint

The following equations (3.1) and (3.2) respectively yield:

And obtaining the flexion and extension angle vectors of the hip joint and the knee joint of the right lower limb in the same way:

wherein a is_x3,a_y3The acceleration components a of the right thigh IMU sensor at the time t after the processing in the step 2 are respectively_x4,a_x4The acceleration degrees of the IMU sensor of the right shank at the time t after the processing in the step 2 are respectivelyAn amount;

finally obtaining the flexion and extension angle vector α of the hip joint and the knee joint of the left lower limb at all K moments^{Right side}And β^{Right side}：

4. Continuous target value real-time prediction for improved LightGBM method

The input of the control signal of the exoskeleton control system directly influences the fluency of the exoskeleton movement, so that the higher the input frequency of the control signal is, the more continuous the signal value input at a single time is, the more natural the movement process of the actuating mechanism of the exoskeleton can be. Based on the principle, when the gait is predicted by using the known LightGBM algorithm, only one joint angle transformation can be predicted at a time, and the exoskeleton execution mechanism is stopped due to the time required for the prediction algorithm to calculate the prediction result and the delay problem of a control signal to the execution mechanism. The conventional LightGBM prediction model cannot meet the real-time prediction function of an actual exoskeleton, so that a gait prediction algorithm based on the LightGBM needs to be improved relatively to be more suitable for the operation of the exoskeleton.

4.1 predictive model training

wherein w is the window width of the sliding window, N is the number of the sliding windows, the predicted length pLen is taken as the step length of the sliding window, and pLen is more than 1; to ensure betterConstructing a target matrix according to the requirement of K- [ (N-1) x pLen + w]Not less than pLen, i.e. ensuring α^{Left side of}Finally, enough angle values are provided to construct predicted values of the pLen predicted lengths;

wherein

where T () is the LightGBM training function,

4.2 parallel prediction of prediction models

wherein,

4.3 according to the same principle of steps 4.1-4.2, the right lower limb hip joint flexion and extension angle parallel prediction model is constructed as follows:

inputting a model:

and (3) outputting a model:

from the characteristic matrix A of the hip joint of the right lower limb_{α right side}、D_{α right side}Analogy to equation (4.3), we derive the pLen prediction models:

the feature vector in the current latest sliding window of the hip joint of the right lower limb is

The prediction of the hip joint flexion and extension angles of the right lower limb at future pLen moments is realized by connecting pLen trained prediction models in a parallel structure:

wherein,

for the hip joint of the right lower limb at the ith moment in the futurePredicting a bending and stretching angle;

the right lower limb hip joint flexion and extension angle prediction vector P can be obtained from the formula (4.5)^{Right side}：

Similarly, the left lower limb knee joint flexion and extension angle parallel prediction model is constructed as follows:

inputting a model:

and (3) outputting a model:

from the characteristic matrix a of the knee joint of the left lower limb_{β left side}、D_{β left side}Analogy to equation (4.3), we derive the pLen prediction models:

the feature vector in the current latest sliding window of the knee joint of the left lower limb is

The left lower limb knee joint flexion and extension angle prediction at future pLen moments is realized by connecting pLen trained prediction models in a parallel structure:

wherein,

predicting a bending and stretching angle of the knee joint of the left lower limb at the ith moment in the future;

the left lower limb knee joint flexion and extension angle prediction vector Q can be obtained from the formula (4.6)^{Left side of}：

Similarly, the right lower limb knee joint flexion and extension angle parallel prediction model is constructed as follows:

inputting a model:

and (3) outputting a model:

from the characteristic matrix a of the knee joint of the right lower limb_{β right side}、D_{β right side}Analogy to equation (4.3), we derive the pLen prediction models:

the feature vector in the current latest sliding window of the knee joint of the right lower limb is

The prediction of the flexion and extension angles of the knee joints of the right lower limbs at future pLen moments is realized by connecting pLen trained prediction models in a parallel structure:

wherein,

predicting a bending and stretching angle of the knee joint of the right lower limb at the ith moment in the future;

the right lower limb knee joint flexion and extension angle prediction vector Q can be obtained from the formula (4.7)^{Right side}：

The step length of each sliding of a sliding window constructed by a data set when gait data is applied to a machine learning algorithm is 1, and the set prediction step length is required to be larger than 1 in order to realize continuous prediction of a target value.

Fig. 3 is a histogram of the contrast between the three algorithms. Compared with the XGboost and LightGBM algorithm based on Gradient Boosting, the gait prediction RMSE of Kalman filtering is higher, but the SC is lower, so that the prediction precision of Kalman filtering is low, but the prediction result is smoother, and the Kalman filtering prediction algorithm does not need training; in both XGBoost and LightGBM based on Gradient Boosting, RMSE of the predicted result is roughly the same overall, but SC of the predicted result of LightGBM is smaller than XGBoost, and training time of LightGBM is significantly smaller than that of XGBoost of the same training set size. Therefore, in the actual application process, the LightGBM can more quickly train a prediction model with good prediction effect.

The inventive prediction model validation is shown in fig. 4.

Step (5) executing the prediction result by the execution mechanism

Obtaining hip joint flexion and extension angles and knee joint flexion and extension angles of the left and right lower limbs according to the step 4 to obtain a prediction track; the left and right lower limb exoskeleton executing mechanisms convert the predicted bending and stretching angle tracks into corresponding control signals through the prior art, and then control motors at corresponding joints to operate so as to realize the gait walking of the outer limbs.

Claims

1. an exoskeleton gait prediction method based on LightGBM, is characterized in that this method comprises the steps:

Step (1), data collection stage

Set the five IMU sensors on the user's waist, left thigh, left calf, right thigh, and right calf respectively; obtain the x and y axial acceleration values of the five IMU sensors in real time;

Step (2), data analysis and denoising and smoothing

According to the x and y axial acceleration values of the five IMU sensors in step (1), a matrix composed of 10×k acceleration components is constructed, which is defined as:

R=[V ₁ ,V ₂ ,...,V _i ,...,V ₁₀ ] ^T , where T represents the transpose of the matrix

in

represents the acceleration value of the i-th acceleration component at time t, and K is the total number of data sampling points;

Perform moving average filtering and denoising processing on V _i , set

The filter value is

but:

Wherein L is the number of original acceleration values taken during the moving average filtering process, and L is an odd number, M=(L-1)/2;

According to formula (2.1), the denoised data of all _Vi at time t (t=1,2,...,K) are finally obtained:

Step (3), obtain the real-time flexion and extension angle in the gait data

According to the de-noised data processed in step (2), taking the left lower limb as an example, at a certain moment t, the hip flexion and extension angle of the left lower limb is

and knee flexion angle

It is obtained by formulas (3.1) and (3.2) respectively:

Wherein a _x1 and a _y1 are respectively the acceleration components of the left thigh IMU sensor at time t after step 2 is processed, and a _x2 and a _x2 are respectively the acceleration components of the left calf IMU sensor at time t after step 2 processing;

Finally, the flexion and extension angle vectors αleft and ^βleft of the ^left lower extremity hip joint and knee joint at all K moments are obtained:

Similarly, the flexion and extension angle vectors of the right lower limb hip joint and knee joint are obtained;

Step (4), real-time prediction of continuous target value of the improved LightGBM method

Take the left lower extremity hip flexion and extension angle α ^left as an example:

4.1 Predictive model training

The vector data of the left lower extremity hip flexion and extension angle obtained through step 3 is a training set, and the prediction model of the left lower extremity hip flexion and extension angle is trained;

First, use the sliding window to extract the angle value from αleft to construct the ^left lower extremity hip joint flexion and extension angle matrix A _αleft , which is used as the feature matrix for training the prediction model, see formula 4.1;

Where w is the window width of the sliding window, N is the number of sliding windows, and the prediction length pLen is used as the step size of the sliding window, pLen>1; in order to ensure better construction of the target matrix, it is necessary to satisfy K-[(N-1)× pLen+w]≥pLen, that is, to ensure that there is enough angle value at the end of α ^left to construct the prediction value of pLen prediction length;

Then, according to the prediction length and the flexion-extension angle value of the ^left middle α relative to the current sliding window in the future, a matrix D _{α left} is constructed as the prediction target value matrix of the training prediction model, see formula 4.2;

in

is the predicted value vector of the left lower extremity hip flexion and extension angle corresponding to the ith moment in the future;

According to formulas (4.1) and (4.2), the feature matrix of the left lower limb knee joint is constructed and substituted into formula (4.3) to obtain pLen prediction models:

where T() is the LightGBM training function,

is the prediction model (function) corresponding to the i-th moment in the future after training on the flexion and extension angle of the left lower extremity hip joint;

4.2 Prediction Model Parallel Prediction

The feature vector in the current latest sliding window of the left lower extremity hip joint is

By connecting pLen trained prediction models in a parallel structure, the prediction of the left lower extremity hip flexion and extension angle at pLen moments in the future is realized:

in,

is the predicted value of the left lower extremity hip flexion and extension angle at the i-th time in the future;

From formula (4.4), the prediction vector Pleft of the ^left lower extremity hip joint flexion and extension angle can be obtained:

Similarly, the parallel prediction model of the flexion and extension angles of the left lower extremity knee joint, right lower extremity hip joint and knee joint is obtained;

Step (5), predict the vector according to step 4 to realize the gait prediction trajectory.